Estimating Advertising Half-life and the Data Interval Bias

We compare three methods of estimating the duration, or half-life, of advertising using computer simulation experiments. In particular, we investigate how well each method works with the data aggregated over different time intervals. In contrast with the existing theory on the, so called, data interval bias, our experiments are based upon realistic advertising schedules. Our results appear to indicate that the indirect "t-ratio" estimation procedure favoured by practitioners works well in the presence of such temporal aggregation. Additionally, we suggest a transformation that can be used in combination with the indirect "t-ratio" estimation procedure to obtain estimates of the underlying micro-period half-life from a variety of common (macro) data frequencies.